Kikuo Maekawa

Balanced corpus of contemporary written Japanese

Abstract The balanced corpus of contemporary written Japanese (BCCWJ) is Japan’s first 100 million words balanced corpus. It consists of three subcorpora (publication subcorpus, library subcorpus, and special-purpose subcorpus) and covers a wide range of text registers including books in general, magazines, newspapers, governmental white papers, best-selling books, an internet bulletin-board, a blog, school textbooks, minutes of the national diet, publicity newsletters of local governments, laws, and poetry verses. A random sampling technique is utilized whenever possible in order to maximize the representativeness of the corpus. The corpus is annotated in terms of dual POS analysis, document structure, and bibliographical information. The BCCWJ is currently accessible in three different ways including Chunagon a web-based interface to the dual POS analysis data. Lastly, results of some pilot evaluation of the corpus with respect to the textual diversity are reported. The analyses include POS distribution, word-class distribution, entropy of orthography, sentence length, and variation of the adjective predicate. High textual diversity is observed in all these analyses.

Coarticulatory reinterpretation of allophonic variation: Corpus-based analysis of /z/ in spontaneous Japanese

Abstract In Standard Japanese, the phoneme /z/ is realized variably either as an affricate or a fricative. This variation was analyzed using a phonetically annotated part of the Corpus of Spontaneous Japanese. Contrary to the traditional linguistic account, the variation is not a positionally conditioned, hence categorical, allophonic variation. Although the effect of linguistic position exists to some extent, its influence is secondary compared to the influence of local temporal characteristics of speech called time allotted for consonant articulation (TACA). The overall prediction rate of the manner of /z/ articulation by means of TACA was 74%, and, when coupled with information on linguistic position, the prediction rate was as high as 80%.

Effects of Focus on Duration and Vowel Formant Frequency in Japanese

The effect of contrastive focus upon duration and vowel formant frequency was examined. The effect upon duration was found at various levels involving utterance, accentual phrase, and individual segments. The effect upon the formant frequencies was also confirmed; it could be different depending on the prosodic location of the vowel in question. Vowels that were directly focussed became more peripheral in terms of vowel height, while /e/ vowels outside the domain of focus became less peripheral in terms of frontness. It was interesting that the observed effects of focus were mostly omni-directional, i.e., focus influenced not only the temporally preceding constituents but also the following ones in all phonetic parameters examined. This requires certain revision of the treatment of focus in the current phonological theory of intonation.

Prosody Discrimination by Songbirds (Padda oryzivora)

In human verbal communication, not only lexical information, but also paralinguistic information plays an important role in transmitting the speakers’ mental state. Paralinguistic information is conveyed mainly through acoustic features like pitch, rhythm, tempo and so on. These acoustic features are generally known as prosody. It is known that some species of birds can discriminate certain aspects of human speech. However, there have not been any studies on the discrimination of prosody in human language which convey different paralinguistic meanings by birds. In the present study, we have shown that the Java sparrow (Padda oryzivora) can discriminate different prosodic patterns of Japanese sentences. These birds could generalize prosodic discrimination to novel sentences, but could not generalize sentence discrimination to those with novel prosody. Moreover, unlike Japanese speakers, Java sparrows used the first part of the utterance as the discrimination cue.

Archiving and Analysing Techniques of the Ultra-Large-Scale Web-Based Corpus Project of NINJAL, Japan

In 2011, the National Institute for Japanese Language and Linguistics (NINJAL) launched a corpus compilation project to construct a web corpus for linguistic research comprising ten billion words by 2016. The project is divided into four categories: Page Collection, Linguistic Annotation, Release and Preservation. For Page Collection, web crawlers are employed to collect web text by crawling 100 million pages every three months and retaining several versions of the text for three-month periods. For Linguistic Annotation, the linguistic studies web corpus contains annotated linguistic information. To improve the usability of these linguistic resources, normalization tasks such as tag removal, word segmentation, dependency parsing, and register estimation are performed. For Release, word lists and n-gram data are published based on the crawled and annotated text corpus. In addition, applications are being developed to enable searching for morphosyntax patterns in the ten-billion-word corpus. For Preservation, crawled web pages are preserved in chronological order as web archives primarily to support the survey of ongoing linguistic changes. In this paper, we present the basic design of the four categories. Additionally, we report the current status of the corpus using basic statistics of the crawled data and discuss the importance of deduplicating sentences.

Voice-Quality Difference Between the Vowels in Filled Pauses and Ordinary Lexical Items.

Acoustic differences between the vowels in filled pauses and ordinary lexical items such as nouns and verbs were examined to know if there was systematic difference of voice-quality. Statistical test of material taken from the Corpus of Spontaneous Japanese showed that, in most cases, there was significant difference of acoustic features like F0, F1, F2, intensity, jitter, shimmer, TL, H1-H2, H1-A2, duration, etc. between the two classes of vowels. Random forest classification of open data sets showed higher than 0.8 Fvalues on average. It turned out intensity, F0, F1, jitter, and H1-H2 were the most important acoustic features for the expected voice-quality difference.

Domain of final lowering in spontaneous Japanese

There are two opposing predictions about the domain of final lowering (FL) in Japanese intonation. One predicts that the domain is the last mora of utterance whereas the other predicts that the domain is much wider (with no clear specification of the domain). X-JToBI annotated part of the Corpus of Spontaneous Japanese (the CSJ-Core, 44 h speech spoken by 201 speakers) was analyzed to determine the domain of FL in Tokyo Japanese. Mean normalized F0 values of the four constituent tones of accented accentual phrases (AP) were compared across utterances differing both in the number of constituting APs (from 1 to 5) and the syntactic strength of sentence- or clause-boundaries (three levels). It turned out that all tones in the last AP of utterance were considerably lowered in all utterances regardless of utterance length. This suggests strongly that the domain of FL is the last AP in Japanese. It also turned out that FL occurred in much wider contexts than hitherto believed. FL was observed not only in typica...

Linguistics-oriented language resource development at the National Institute for Japanese Language and Linguistics

The aim of this talk consists in the introduction to the language-resource-related activities of the National Institute for Japanese Language and Linguistics (NINJAL). Since the last half of the 1990s, the former National Language Research Institute (NLRI) played a central role in the development of Japanese language resources by constructing corpora like Corpus of Spontaneous Japanese (CSJ) and Taiyo Corpus. In 2006, the language resource group of NLRI started a Japanese corpus compilation initiative named KOTONOHA, and set about the construction of a 100 million words Balanced Corpus of Contemporary Written Japanese (BCCWJ). The activity of NLRI was inherited by the NINJAL Center for Corpus Development reestablished in 2009. Now that the construction of the BCCWJ was completed successfully in August 2011, the NINJAL center set about two new projects of exploratory nature: a historical corpus project and a 10-billion-word ultra-large-scale Web-based corpus project. In addition to the presentation of the NLRI-NINJAL activities, language resource development in Japanese institutions other than NINJAL will be introduced briefly in the beginning. Also, application of the CSJ to the study of phonetics will also be demonstrated at the end.

Experimental paradigm influence subject’s perception of attitudes

Japanese utterances spoken with different attitudes (admiration, suspicion, and disappointment) were independently manipulated for pitch‐contour and voice quality using STRAIGHT {Kawahara}. Close copy stylization of the prototypical pitch contour for the three attitudes was imposed on the naturally spoken utterances, producing stimuli with all combinations of voice quality types and pitch‐contour shapes. The utterances were submitted to two separate experiments, wherein subjects were asked to judge the attitude of the morphed utterances. The first, a forced choice experiment, asked subjects to choose if the utterances were admiration, suspicion, or disappointment, and the second, a free choice experiment, where subjects could freely choose the attitude they perceived. The results from the forced choice test indicated that subjects used pitch contour cues to choose speaker attitude. However, the results from the free choice paradigm indicated that subjects used both voice quality and intonation cues, and w...